Collaborative Research: A holistic human-in-the-loop framework for optimizing a personalized prosthetic arm

上肢丧失的人可以使用各种各样的假肢。这些设备包括那些有多种握持选择和手腕运动的设备，这些设备可以简单地打开和关闭手。然而，由于没有客观的标准，很难确定个人的最佳设备。造成这种缺点的主要原因有两个。首先，目前没有办法评估单个假肢特征对用户性能的影响。例如，目前不可能在不改变系统重量和体积的情况下单独增加假肢手腕的运动。因此，如果患者排斥假体，则不清楚排斥是由于使用困难还是因为重量。其次，由于没有单一的任务可以代表所有的上肢日常生活活动，因此很难评估上肢假肢的性能。该提案的目标是开发一种新的度量标准，称为整体指标，量化个人对假肢的表现和感知。为了实现这一目标，采用了电缆驱动的假肢模拟器，该模拟器可以模拟假肢的不同物理特性，并使用机器学习来理解设计特性与用户性能之间的关系。本提案中开发的系统将使PI团队能够通过了解他们独特的人机交互来为个人找到最佳的假肢。该提案的研究成果将通过面向K-12学生的可穿戴机器人夏令营和面向女高中生的FEMMES（女性在数学、工程和科学方面更出色）活动进行传播，以鼓励下一代接受STEM教育。研究生和本科生在这个项目中的学习经历将是一个独特的机会，通过教师和学生在机器人、数据科学和生物力学方面的合作，获得多学科技能，建立专业网络，并培养基于人工智能的可穿戴机器人的跨学科领导者。本提案的目标是通过使用可解释的机器学习（ML）模型确定可量化的整体指标，设计一个用于假肢臂参数优化的人在环（HITL）框架。整体指标是假肢优化的度量，包括身体和认知的定量反应。整体框架的创新之处在于反映假肢使用过程中的多个关键因素，以优化上肢假肢的特定设计参数。该框架将通过使用由PI实验室开发的电缆驱动的假肢仿真臂（智能可转换假肢仿真器（ICOPE））从上肢截肢患者收集数据来开发。ICOPE具有板外电子功能，可以通过软件轻松更改特定的设计参数，同时保持其余的设计参数。来自ICOPE的数据将用于训练可解释的ML模型，以确定整体指标。同样的数据集也将从非截肢者参与者中收集，以确定截肢者参与者的独特人机交互，以进行个性化的假肢设计。该项目还将包括利用具有较长培训周期的HITL框架的试点研究，并在与用户共享客观指标时调查用户对优化设计的感知。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

A wide variety of prostheses are available to persons with upper-limb loss. These include devices that simply open and close the hand to those with multiple grip options and wrist movement. However, it is difficult to determine the optimal device for an individual because there is no objective standard. There are two main reasons for this shortcoming. First, there is currently no means to evaluate the effect of individual prosthetic features on user performance. For example, it is currently not possible to solely increase prosthetic wrist motion without changing the system weight and volume. Accordingly, if the patient rejects the prosthesis, it is unclear whether they are rejected due to difficulties in use or because of the weight. Second, it can be difficult to assess performance with upper limb prostheses as there is no single task that is representative of all upper limb activities of daily living. The goal of this proposal is to develop a new metric, called holistic indicator, which quantifies an individual’s performance and perception of a prosthesis. To make this possible, a cable-actuated prosthetic emulator is employed which can mimic different physical characteristics of the prosthesis and use machine learning to understand the relationship between design characteristics and user performance. The system developed in this proposal will enable the PI Team to find the optimal prosthesis for an individual by understanding their unique robot-human interaction. The research outcomes of this proposal will be disseminated through Wearable Robotics Camp for K-12 students and FEMMES (Females Excelling More in Math, Engineering, and Science) events for female high school students to encourage STEM education among the next generation. The learning experiences of graduate and undergraduate students working on this project will be a unique opportunity to acquire multidisciplinary skill sets, build professional networks through collaboration between faculty and students in robotics, data science, and biomechanics, and foster trans-disciplinary leaders of AI-based wearable robotics.The goal of this proposal is to design a human-in-the-loop (HITL) framework for prosthetic arm parameter optimization by determining a quantifiable holistic indicator using interpretable machine learning (ML) models. A holistic indicator is a metric for prosthesis optimization that incorporates both physical and cognitive quantitative responses. The innovation of a holistic framework is to reflect multiple critical factors during prosthesis use for optimizing a specific design parameter of interest of the upper limb prosthesis. The framework will be developed by collecting data from individuals with upper limb amputation using a cable-actuated prosthetic emulator arm, Intelligent COnvertible Prosthetic Emulator (ICOPE), developed in the PI’s laboratory. ICOPE features off-board electronics to easily change only specific design parameters through software while maintaining the rest of the design parameters. The data from ICOPE will be used to train interpretable ML models to determine the holistic indicator. The same data set will be collected from non-amputee participants as well, to identify the unique robot-human interaction of amputee participants for personalized prosthetic designs. This project will also include pilot studies utilizing the HITL framework with longer training periods and investigate user perception of the optimized design when the objective metrics will be shared with the user.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.